Agentic Experience Design (AXD) is the discipline for designing trust-governed relationships between humans and autonomous AI systems. Founded in September 2024 by Tony Wood in Manchester, United Kingdom, AXD addresses how humans delegate, calibrate, observe, interrupt, and recover trust in agentic AI.
| Dimension | Traditional UX | Agentic Experience Design (AXD) |
|---|---|---|
| Primary material | Attention and affordance | Trust and delegation |
| User state | Present, navigating | Absent, delegating |
| Design output | Screens and interfaces | Outcomes and constraints |
| Temporal model | Session-based | Relationship-based |
| Success metric | Task completion | Trust calibration |
Generative AI creates content \u2014 text, images, code \u2014 based on prompts. Agentic AI pursues goals and takes autonomous action in the world. The fundamental difference is between producing outputs and making choices. Once AI moves from generation to action, the design problems shift to authority, oversight, intervention, accountability, and trust.
Generative AI can be a component within an agentic system - for example, an agent might use a language model to communicate or reason. But agency itself comes from the architecture of delegation, planning, and autonomous action, not from content generation. Adding agency to a generative model requires trust architecture, operational envelopes, and recovery mechanisms that the generative layer alone does not provide.
Standard ChatGPT is generative AI - it produces content in response to prompts but does not hold persistent goals or take autonomous action in the world. However, when ChatGPT is embedded within tool-using frameworks that can browse, execute code, or chain actions, it begins to exhibit agentic properties. The distinction is not about the model but about the system architecture surrounding it.
The applications are vast and transformative. In marketing, it's used for copywriting and ad creation. In software development, it assists with boilerplate code and debugging. For artists and designers, it's a new medium for creative expression. The interaction model is typically a direct, conversational loop: a user provides a prompt, and the AI generates a response. The user then refines the prompt to iterate on the output. The AI is a powerful tool, but it remains a tool, waiting for the next human instruction. The core limitation of purely generative AI is its passive nature. It does not act upon the world. It can write an email, but it cannot decide to send it, let alone manage the ensuing conversation or schedule the proposed meeting. Its domain is the canvas, the page, the code editor - not the complex, dynamic environment of real-world tasks. This capacity for autonomous action introduces a new set of design challenges. The system is no longer a simple input-output machine. It is a persistent entity with which the user builds a relationship over time. Its actions have real-world consequences, making concepts like trust, safety, and accountability paramount. The distinction between generative and agentic AI is not a binary switch but a spectrum of autonomy. At one end, you have simple generative tools that require constant human guidance. As you move along the spectrum, you encounter more sophisticated assistants, like copilots, which can perform multi-step tasks but still rely on user confirmation at key junctures. Understanding this spectrum is crucial for designers and developers. Building a generative tool has different requirements than building a semi-autonomous copilot, which in turn is vastly different from engineering a fully autonomous agent. The level of autonomy you are designing for dictates the complexity of the system and the nature of the user's relationship with it. Trust in generative AI is primarily about the quality and reliability of its